Vals AI raised a $40 million Series A at a $400 million valuation as the company scales an independent AI evaluation platform that measures how well frontier models perform on real-world professional tasks. Andreessen Horowitz (a16z) led the round, with participation from existing investors 8VC and Bloomberg Beta and new investors HRT Ventures and Next Ladder Ventures.
The financing comes as Vals AI expands rapidly. The company said its revenue has grown eightfold compared with all of 2025, its customer base has doubled, and its team has tripled over the past six months.
Vals AI was founded around the thesis that the speed of artificial intelligence development is beginning to exceed the industry’s ability to independently and reliably measure progress.
AI developers have historically improved models through a process resembling hill climbing, in which researchers define measurable objectives, identify weaknesses and repeatedly optimize against those benchmarks.
But Vals argues that increasingly capable models can rapidly saturate existing benchmarks, while benchmark questions can also leak into model training data and diminish the usefulness of those tests.
Another concern is that many AI benchmarks are developed or operated by the same organizations building the models being evaluated.
Vals AI is positioning itself as an independent measurement layer between model developers and organizations deciding how AI should be deployed.
The company builds benchmarks that evaluate whether AI models can perform work traditionally completed by professionals, including lawyers, bankers, engineers, and doctors.
Creating those evaluations involves working with reference institutions in each field to develop taxonomies of representative tasks and methodologies for automatically evaluating the quality of AI-generated work.
Vals also maintains private test sets intended to preserve benchmark integrity by preventing models from being directly trained on the evaluation material.
The company has developed infrastructure that lets evaluations run reproducibly and at scale across different AI labs, and has released portions of it as open source.
Vals’ results have been cited in model cards from OpenAI, Anthropic, Google, Meta and xAI.
The platform is also being used by major enterprise AI deployments to determine which models to build on and to compare their own products against frontier systems.
Government is becoming another important part of the company’s opportunity.
Vals said it has supported the U.S. Department of Commerce and members of Congress working on AI policy as policymakers develop ways to measure frontier AI capabilities, cybersecurity risks and international technological competition.
The company’s broader argument is that AI is becoming a massive economic market without the independent measurement institutions that exist in sectors such as finance and healthcare.
For AI developers, better independent evaluations can provide evidence that newer models represent genuine improvements.
For enterprises, benchmarking can help determine whether investments in AI produce measurable returns and which models are best suited for specific business functions.
And for governments, independent testing can provide information about model capabilities and emerging risks without relying entirely on assessments produced by AI developers themselves.
Alongside the Series A announcement, Vals introduced three major product and benchmark initiatives.
Vals Smith is now generally available and allows users to create customized coding benchmarks from GitHub repositories.
The company is initially providing 120 free credits for users building benchmarks with the product.
Vals is also expanding its Frontier Risk Benchmarks.
The new work includes an RSI Index developed in collaboration with CoreWeave, a cybersecurity benchmark created with academic researchers and initial evaluation work focused on mental health.
The company has also launched Vals 2.0, which includes a rebuilt website and a new Vals Index designed to provide broader coverage of AI performance across the economy.
With the $40 million Series A and $400 million valuation, Vals AI is positioning independent AI measurement as infrastructure that could become increasingly important as model capabilities advance and businesses and governments seek more objective ways to determine what those systems can actually do.
